gulltoppr
Enables AI agents to resolve smart contract ABIs, read, encode, simulate, and prepare transactions across multiple blockchains via a REST API or MCP server, with no signing required.
README
gulltoppr
The REST engine for abi.ninja-for-agents — the resolution ladder + verb surface
that lets an AI agent go from (chain, address) to a correct, simulated, safe
contract interaction. This is "the engine" of the four faces (REST → MCP → SDK →
Skill); see ../SPEC.md for the full contract and ../IDEATION.md
for the strategy.
TypeScript + viem + Hono. The heimdall decompile rung is delegated over HTTP to gulltoppr (kept out-of-process by design).
Run
npm install
npm run dev # REST engine — tsx watch on http://localhost:8787
npm run mcp # MCP server — stdio, 7 tools (for agent clients)
npm run mcp:http # MCP server — Streamable HTTP (remote agents)
npm run typecheck # tsc --noEmit
npm test # vitest — unit tests (cache, chains, ladder helpers, args, errors)
Env
| var | default | notes |
|---|---|---|
PORT |
8787 |
|
HEIMDALL_API_URL |
https://heimdall-api.fly.dev |
heimdall decompile service (ladder rung 4) |
ETHERSCAN_API_KEY |
(empty) | one multichain v2 key; empty disables rung 1 |
SIGNING_BASE_URL |
https://abi.ninja |
base for prepare_tx hand-off deeplinks |
RATE_LIMIT |
120 |
per-IP requests per window (fixed window); 0 disables |
RATE_LIMIT_WINDOW_SEC |
60 |
rate-limit window length |
RATE_LIMIT_ALLOW |
(empty) | comma-separated IP allowlist (exempt); private 6PN IPs are always exempt |
ANTHROPIC_API_KEY |
(empty) | enables the registry's LLM propose-and-verify pass on decompiles; empty disables |
REGISTRY_LLM_MODEL |
claude-opus-4-8 |
model for propose-and-verify |
Endpoints (SPEC §4)
| verb | route |
|---|---|
resolve_abi |
GET /v1/{chain}/{address}/abi |
read_contract |
POST /v1/{chain}/{address}/read — {function, args} |
encode_call |
POST /v1/{chain}/{address}/encode — {function, args, value?} |
simulate |
POST /v1/{chain}/simulate — {from,to,data,value?} or {from,address,function,args,value?} |
prepare_tx |
POST /v1/{chain}/{address}/prepare — {function, args, from, value?} |
decode_tx |
GET /v1/{chain}/tx/{hash} |
resolve_name |
GET /v1/{chain}/name/{name} · GET /v1/{chain}/name/by-address/{address} |
| registry lookup | GET /v1/lookup/{selector} — 4-byte (function/error) or 32-byte (event topic0), chain-independent |
| registry stats | GET /v1/registry/stats |
The registry (selector commons)
The engine seeds an open selector→signature registry as a byproduct of resolution:
- Every verified resolution (Etherscan/Sourcify) harvests ground-truth
selector → signaturepairs for functions, events (full 32-byte topic0 — collision-free), and errors. Proof grade:verified-source. - Resolutions are also indexed by skeleton hash (runtime bytecode with the
solc metadata trailer stripped), so byte-identical clones resolve via a new
bytecode-matchrung without re-running the ladder. Verified claims are capped topartialfor clones (this address's source was never verified). - Decompiled ABIs get
Unresolved_<selector>names replaced from proven registry entries, and (whenANTHROPIC_API_KEYis set) a fire-and-forget propose-and-verify pass asks Claude for candidate signatures and accepts only those wherekeccak256(sig)[:4]reproduces the selector — proof gradekeccak-proven(signature proven; semantics still inferred).
Only the engine's own pipeline writes to the registry — no open submissions (that's how 4byte got collision-poisoned).
{chain} is an alias (ethereum, base, optimism, arbitrum, polygon,
local) or a numeric id. Pass ?rpc_url= to override the RPC (required for chains
with no default, e.g. local/31337).
curl localhost:8787/v1/ethereum/0xC02aaA39b223FE8D0A0e5C4F27eAD9083C756Cc2/abi
curl -X POST localhost:8787/v1/ethereum/0xC02aaA39b223FE8D0A0e5C4F27eAD9083C756Cc2/prepare \
-H 'content-type: application/json' \
-d '{"function":"approve","args":["0x1111111254EEB25477B68fb85Ed929f73A960582","1000000000000000000"],"from":"0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045"}'
MCP server (SPEC §5)
npm run mcp starts a stdio MCP server exposing the same seven verbs as tools. The
tools are a thin adapter over the deployed REST engine via gulltoppr
(ENGINE_URL), so the MCP shares the engine's persistent cache and Etherscan key —
no duplicated resolution or secrets. Tool descriptions bake in the non-custodial
hand-off model (prepare_tx never signs) and lead with provenance warnings when an
ABI is decompiled.
Wire it into an MCP client (Claude Desktop / Claude Code mcp config):
{
"mcpServers": {
"gulltoppr": {
"command": "npm",
"args": ["run", "--silent", "mcp"],
"cwd": "/home/ubuntu/repos/abi-agent",
"env": { "ETHERSCAN_API_KEY": "" }
}
}
}
Tools: resolve_abi, read_contract, encode_call, simulate, prepare_tx,
decode_tx, resolve_name. All are read-only-annotated except prepare_tx
(non-destructive — returns an unsigned hand-off, signs nothing).
Remote (Streamable HTTP)
For agents that can't run a local stdio server, the same MCP is hosted over HTTP at
https://gulltoppr-mcp.fly.dev/mcp (npm run mcp:http locally; stateless). Point
an HTTP-capable MCP client at that URL:
{ "mcpServers": { "gulltoppr": { "url": "https://gulltoppr-mcp.fly.dev/mcp" } } }
Tool registration is shared (src/mcp-server.ts) between the stdio entry (mcp.ts)
and the HTTP entry (mcp-http.ts), deployed via Dockerfile.mcp / fly.mcp.toml.
npm SDK
A typed client over this REST surface lives in sdk/ (gulltoppr) —
new AbiNinja({ baseUrl }).resolveAbi(...) / .read(...) / .prepareTx(...), plus
a contract() helper. It's the third face (after REST and MCP) and the basis for
refactoring abi.ninja's frontend onto a shared client. See sdk/README.md.
Deploy
Live at https://gulltoppr.fly.dev (Fly.io, region cdg — co-located with
gulltoppr to minimize ladder rung-4 latency). Containerized via the Dockerfile
(Node 22, run with tsx; ~82 MB image), configured by fly.toml.
flyctl deploy --remote-only --ha=false
# optional: set an Etherscan v2 key to enable ladder rung 1
flyctl secrets set ETHERSCAN_API_KEY=... -a gulltoppr
HEIMDALL_API_URL / SIGNING_BASE_URL / PORT are set in fly.toml [env].
Machines auto-stop when idle and auto-start on request.
Claude Skill
The fourth face: a Claude Skill (skill/gulltoppr/) that teaches an agent
the workflow — resolve → check provenance → read or prepare → simulate → hand off —
and the non-custodial safety rules. Install with
cp -r skill/gulltoppr ~/.claude/skills/gulltoppr. See skill/README.md.
Layout
src/
server.ts REST routes (Hono), BigInt-safe JSON, error mapping
index.ts REST entry / boot
mcp.ts MCP server (stdio) — 7 tools over the same verbs
config.ts env + defaults
chains.ts alias/id → {id, viem chain, rpc} (SPEC §6)
clients.ts cached viem PublicClients
types.ts the SPEC §2 data types
errors.ts typed ApiError → HTTP status (SPEC §7)
resolve/
index.ts resolve_abi — the ladder orchestrator (the spine)
etherscan.ts rung 1 · sourcify.ts rung 2 · proxy.ts rung 3
heimdall.ts rung 4 (gulltoppr) · fourbyte.ts rung 5
interface.ts capability manifest builder ("the buttons", SPEC §2.4a)
selectFunction.ts name/signature → AbiFunction
verbs/
read.ts encode.ts simulate.ts prepare.ts decodeTx.ts resolveName.ts
args.ts JSON-arg → viem-typed coercion
Status
Working end-to-end (verified against live mainnet): the full ladder, the
capability manifest, read_contract, encode_call, prepare_tx (with eth_call
simulation + deeplink + provenance warnings), decode_tx (via gulltoppr), and ENS
resolve_name — exposed over both the REST surface and the MCP server
(stdio handshake + all 7 tools + a live tool call verified).
Stubbed / TODO (clearly marked in-code):
- 4byte rung 5 — returns null; ladder ends in
ABI_NOT_FOUNDinstead of a selector-only ABI. Needs bytecode selector scan + 4byte.directory lookup. simulatestate_diff — empty; needsprestateTracer.asset_changes/logscome fromdebug_traceCall(callTracer) when the RPC supports it, else empty.- basenames (
*.base.eth) —resolve_nameonly does mainnet ENS today. - diamonds (EIP-2535) — proxy detection covers 1967/UUPS/transparent/beacon/1167.
decode_tx— doesn't yet layer a verified ABI over the heimdall decode for real event/param names.- caching — no result cache yet; every
resolve_abire-runs the ladder (gulltoppr caches its own decompiles).
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。